Fine Registration of SAR and Optical Image Based on Improved Hausdorff Distance

Abstract

The registration of SAR and optical image is a core step of multi-source image fusion. The quality of registration restricts the effect of multi-source image fusion. Due to the importance of SAR and optical image registration, we analyze the difficulties of SAR and optical image registration first. And we propose a fine registration method for SAR and optical image based on improved Hausdorff distance. Firstly, we correct the nonlinear distortion of SAR image in range direction by interpolation. Secondly, we use the genetic algorithm to search the mapping parameters that minimize the Hausdorff distance between the images to be registered. These parameters are used for affine transformation of optical images to achieve coarse registration. Finally, we use the Fourier operator to calculate the residual translation between SAR and optical image. Experiments on real data show the effectiveness of the proposed method, and the offset between the corresponding images is less than 2 pixels.